Embed Size (px)
Citation preview
EFFECTS OF TALKER IDENTITY ON SPEECH PROCESSING: �
A LIFESPAN PERSPECTIVE
London 30.3.2020
Adriana HanulíkováUniversity of Freiburg
Throughoutthelifespan,weacquireknowledgeaboutspeakers,theiriden88esandtheirideasandthatcanleadtoexpecta8onsandstereotypes.Previousresearchhasshownthatsuchexpecta8onscanposi8velyornega8velyinfluencelanguageprocessing.InthistalkIexaminehowsuchexpecta8onsaffectspeechintelligibilityandspeechevalua8on.Iwillfocusonexpecta8onsdrivenbyperceivedethnicity.
1. How does perceived ethnicity affect accent ratings and speech intelligibility in different speech contexts?
2. Do talker effects change over the lifespan and across different tasks and speech contexts?
Thankyoufororganizingthisworkshop.
e.g. Rubin, 1992; Kang & Rubin, 2009, Babel & Russel, 2015; Zheng & Samuel, 2017; Hanulíková, 2018)
e.g. McGowan, 2015; Fiedler, Keller, Hanulíková 2019; Hanulíková, in prep.)
Senkyoufororganizingzisvorkshop.
©AdrianaHanulíková London2020 2
Previousstudiesreportconflic8ngresults.Incontrasttonon-na8vespeech,whenpresen8ngEnglishspokenbyana8vespeaker,intelligibility,comprehensionscores,andaccentra8ngsdropforanAsiancomparedtoaWhiteCaucasian/Europeanguise.Thisparadoxisexplainedintermsofmatchingexpecta8ons:Whentheexpectedandobservedspeechmatch,intelligibilityisenhanced.However,differentstudiesusedifferentmeasuresandonespeechcontextonly.HereIcomparethreecontextsandtwomeasuresacrossnon-universitypopula8ons.
Social attributions & linguistic expectations
Theoretical accounts and predictions• Expectation/socially indexed exemplar-based accounts:
anticipating a talker’s accent can enhance intelligibility– Intelligibility of foreign-accented speech, but not regional and
native speech, enhanced for Asian relative to white European primes
– Accent ratings change as a result of varying expectations • Bias-based accounts/reverse-linguistic stereotyping:
conscious misunderstanding of the speech signal or decrease in perceptual effort in the case of a speaker classification as nonnative (stereotypical attributions)– Intelligibility and accent ratings drop for the Asian compared
to the white European primes in all accent conditions ©AdrianaHanulíková London2020 3
Therearetwomaincompe8ngtheore8calaccountsthatmakedifferentpredic8ons.
Talker effects across the lifespan• Intelligibility of speech declines with age – overall drop in performance for elderly
• World knowledge increases with age (e.g., Park et al. 2002)
• Stereotypical inferences increase with age (e.g. Radvansky et al., 2010)
– more pronounced talker effects on intelligibility and accent ratings with increasing age
©AdrianaHanulíková London2020 4
Whileabundantresearchexiststhatdealswithdeclinesinspeechprocessingskills,lessisknownabouttalkereffectsacrossthelifespan.
Experimental design
Stimuli: 36 unpredictable utterances (from Levy et al. 2019) in speech-shaped noise, three speech contexts: standard, regional accent, foreign accent (Korean), and 6 visual primes (from the Chicago Database)Participants: Teens à 72, mean age 14.1; range 12-15
Adults à 50, mean age 36; range 30-45Elderly à 50, mean age 77.6; range 70-92
©AdrianaHanulíková London2020 5
DeralteHelmangelt.(theoldhelmetisfishing)
Task:writedown/repeatwhatthespeakersaid
DieblaueKatzeredet.(thebluecatistalking)
Task:writedown/repeatwhatthespeakersaid
Ra8ngs
Ra8ngs
Similardesigntopreviousstudies:SubjectssawapictureofthespeakerandheardaGermanu\erance.Theythenwrotedown/repeatedtheu\erance,ra8ngswerecollectedattheend.
foreign regional standard
AccentAc
cent
edne
ss ra
tings
(5 =
ver
y st
rong
)
12
34
5
Asian faceWhite European face
Results: accentedness ratings TEENS ADULTS
foreign regional standard
Accent
Acce
nted
ness
ratin
gs (5
= v
ery
stro
ng)
12
34
5
AsianWhite European
• Stereotypical attributions lead to negative ratings• Differences in ratings across the lifespan: greater for the elderly
©AdrianaHanulíková London2020 6
foreign regional standard
Accent
Acce
nted
ness
ratin
gs (5
= v
ery
stro
ng)
12
34
5
AsianWhite European
foreign regional standard
Accent
Acce
nted
ness
ratin
gs (5
= v
ery
stro
ng)
12
34
5
AsianWhite European
ELDERLY
Cumula8velinkmixedmodelacrossallgroups:maineffectofFace,noeffectofGroupandnointerac8ons.Withineachgroup:maineffectofFaceforteensandelderlyonly.
ALL GROUPS
• Effect of talker identity on intelligibility in non-native speech only, and greater for teens
©AdrianaHanulíková London2020 7
foreign regional standard
Accent
Prop
ortio
n co
rrect
ly re
peat
ed w
ords
0.0
0.2
0.4
0.6
0.8
1.0Asian faceWhite European face
foreign regional standard
Accent
Prop
ortio
n co
rrect
ly re
peat
ed w
ords
0.0
0.2
0.4
0.6
0.8
1.0Asian faceWhite European face
foreign regional standard
Accent
Prop
ortio
n co
rrect
ly re
peat
ed w
ords
0.0
0.2
0.4
0.6
0.8
1.0Asian faceWhite European face
foreign regional standard
Accent
Prop
ortio
n co
rrect
ly re
peat
ed w
ords
0.0
0.2
0.4
0.6
0.8
1.0Asian faceWhite European face
TEENS ADULTS
ELDERLY
*
Acrossallgroups:maineffectofGroup,interac8onsbwFaceandAccentandAccentandgroup,butno3-wayinterac8onwithGroup.
ALL GROUPS
Results: intelligibility
*
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
Adjustement?
Prop
ortio
n co
rrec
tly re
peat
ed w
ords
©AdrianaHanulíková London2020 8
ALL GROUPS
foreign regional standard
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
first second0.0
0.2
0.4
0.6
0.8
1.0congruentincongruent
*
• Listeners adjust their expectations; talker effects short-lived
Plotsnowshowhowcongruencyeffectchangesinthefirstandsecondhalfoftheexperiment.Intheforeigncondi8on,congruencereferstotheAsianguise,intheregionalandstandardcondi8on,congruencereferstotheWhiteEuropeanguise.
Theregionalaccentwashardtounderstand.Par8cipantsini8allythoughtthatitisnon-na8vespeech,thecongruenceeffectisthereforedelayedandemergesinthesecondhalfonly.
Summary1: Talker effects in different accents and tasks?
• Greater difficulties with distorted speech for elderly relative to teens and adults
• Intelligibility: talker effects less pronounced with age
• Accentedness ratings: talker effects more pronounced for elderly
• Intelligibility: perceived ethnicity can enhance speech intelligibility, but effects vary for different speech contexts
• Talker effects most pronounced in the foreign and partly regional accents
• Accentedness ratings: more negative for Asian primes irrespective of the speech context
2: Changes in talker effects across the lifespan?
©AdrianaHanulíková London2020 9
Discussion• Expectations/socially indexed exemplars vs.
Biases/reverse-linguistic stereotyping– Intelligibility• of the foreign accent enhanced for the Asian primes• of the regional accent reduced for the Asian primes
(second half only)• of the native speech not affected by talker ethnicity
– Accentedness ratings• Overall more negative for the Asian primes compared
to the white European primes
©AdrianaHanulíková London2020 10
✔✗✔✔
✗ ✗
✗✔
Theintelligibilityresultsarebestaccountedforbyexpecta8onaccounts,butpostperceptualofflinetasksuchasra8ngsaremorepronetobiases.Boththeoriescanaccountforsomeofthefindings.Suchadisconnectbetweenlinguis8cmeasuresisnotnovel,butweneedtobeawareofthesedifferences,becausetheyobscurecomparisonsofdifferenttheories.
�Feel free to send a question related to
this research. �([email protected])
• Thanks to:• Personal project grant of the Ministry of Science, Research and Art,
Baden-Württemberg• Psycholinguistics Lab, University of Freiburg: �
Dr. Daniel Müller-Feldmeth, Dina Osime, Saskia Leymann, Lisa Hübner• Frias project group grant: Language Dynamics across the Life Span, with
Prof. Evelyn Ferstl, Dr. Alice Blumenthal-Dramé
©AdrianaHanulíková London2020 11
